System and method for generating queries by machine learning models

Abstract

A method and system for generation of queries by machine learning (ML) models is provided. An ML model may generate a query based on reception of a data trigger. The query may be generated based on corresponding domain and a knowledge graph. The ML model may receive a response in an encoded format. The knowledge graph may evolve based on the response. A first subsequent query may be generated by the ML model based on the response and the evolved knowledge graph. The ML model may receive a response for the subsequent query in the encoded format. The ML model may determine whether the response culminates a current iteration. A second subsequent query may be generated by the ML model when the response does not culminate the current iteration. The current iteration may be terminated when it is determined that the response culminates the current iteration.

We claim: 
     
       1. A method for generating queries by Machine Learning (ML) models, the method comprising:
 generating, by an ML model, a query upon receiving a data trigger associated with an event, based on a domain of the event and a dynamic knowledge graph associated with the domain; and 
 receiving, by the ML model, a response corresponding to the query in a pre-defined encoded format, wherein the dynamic knowledge graph evolves based on the response; 
 iteratively performing:
 generating, by the ML model, a subsequent query based on the response received for the query and the evolved dynamic knowledge graph, wherein the query immediately precedes the subsequent query; 
 receiving, by the ML model, a response for the subsequent query in the pre-defined encoded format; 
 determining, by the ML model, whether the response culminates the current iteration; and 
 performing, by the ML model, one of: 
 generating a second subsequent query succeeding the subsequent query, when the response does not culminate the current iteration; and 
 terminating the iteration when the response culminates the current iteration. 
 
 
     
     
       2. The method of  claim 1 , wherein the dynamic knowledge graph evolves by dynamically generating nodes within the dynamic knowledge graph, based on each response. 
     
     
       3. The method of  claim 1 , wherein each query generated by the ML model is one of an event specific query and a pseudo query based on a pre-defined threshold value, and wherein each query corresponds to a close ended query generated based on the evolved dynamic knowledge graph. 
     
     
       4. The method of  claim 3 , wherein each query is generated by determining a relevancy score associated with a plurality of path connecting a set of nodes present within the dynamic knowledge graph, and wherein the relevancy score is determined based on a plurality of domain specific parameters. 
     
     
       5. The method of  claim 3 , wherein generating the event specific query comprises: selecting a path from the plurality of paths with the relevancy score greater than the pre-defined threshold value. 
     
     
       6. The method of  claim 3 , wherein generating the pseudo query comprises selecting a path from the plurality of paths with the relevancy score less than the pre-defined threshold value. 
     
     
       7. The method of  claim 1 , wherein the dynamic knowledge graph is one of a Graph Convolutional Network (GCN) and a Generative Adversarial Network (GAN). 
     
     
       8. A system for generating queries by Machine Learning (ML) models, the system comprising:
 a processor; and 
 a memory communicatively coupled to the processor, wherein the memory stores processor-executable instructions, which, on execution, causes the processor to:
 generate a query upon receiving a data trigger associated with an event, based on a domain of the event and a dynamic knowledge graph associated with the domain; and 
 receive a response corresponding to the query in a pre-defined encoded format, wherein the dynamic knowledge graph evolves based on the response; 
 iteratively perform:
 generating a subsequent query based on the response received for the query and the evolved dynamic knowledge graph, wherein the query immediately precedes the subsequent query; 
 receiving a response for the subsequent query in the pre-defined encoded format; 
 determining whether the response culminates the current iteration; and 
 performing one of: 
 generating a second subsequent query succeeding the subsequent query, when the response does not culminate the current iteration; and 
 terminating the iteration when the response culminates the current iteration. 
 
 
 
     
     
       9. The system of  claim 8 , wherein the dynamic knowledge graph evolves by dynamically generating nodes within the dynamic knowledge graph, based on each response. 
     
     
       10. The system of  claim 8 , wherein each query generated by the ML model is one of an event specific query and a pseudo query based on a pre-defined threshold value, and wherein each query corresponds to a close ended query generated based on the evolved dynamic knowledge graph. 
     
     
       11. The system of  claim 10 , wherein each query is generated by determining a relevancy score associated with a plurality of path connecting a set of nodes present within the dynamic knowledge graph, and wherein the relevancy score is determined based on a plurality of domain specific parameters. 
     
     
       12. The system of  claim 10 , wherein, to generate the event specific query, the processor-executable instructions further causes the processor to: select a path from the plurality of paths with the relevancy score greater than the pre-defined threshold value. 
     
     
       13. The system of  claim 10 , wherein, to generate the pseudo query, the processor-executable instructions further causes the processor to: select a path from the plurality of paths with the relevancy score less than the pre-defined threshold value. 
     
     
       14. The system of  claim 8 , wherein the dynamic knowledge graph is one of a Graph Convolutional Network (GCN) and a Generative Adversarial Network (GAN).